Root cause analysis during process development using Joint-Y PLS

The original purpose of the JYPLS model was to aid product transfer and scale-up. This work demonstrates that the JYPLS model is also well suited to analyze data from a process in development where the number and type of sensors and sensor locations are also being decided. The specific application in this case was to determine the root cause for a bias found during the development of a multivariate calibration model for a Near Infrared (NIR) instrument. The calibration model was built in parallel with the development of the manufacturing process itself. Each time the calibration model was tested with data from a new batch run, a non-explained bias was observed. Laboratory results on samples gathered from the process show that the NIR signal itself was biased and not the actual concentration of the constituent. Data on the processing conditions were collected for all the batches and included in the JYPLS model along with NIR and lab results. The loadings and scores from the JYPLS model were interpreted to isolate the root cause of the observed drift in the calibration model. This work discusses the general approach and presents the sequence of diagnostics used to analyze the different campaigns during process development. This should be applicable to other scenarios in process development where the data has the same architecture.